Privacy policy

Key points

• This website uses no cookies, no analytics and no third-party content. Fonts are served from our own server.
• For live captions, audio is sent encrypted directly to the speech service Soniox and turned into text on its EU servers. Our own servers do not receive audio in this process. The exception is a non-public test mode for our own online meetings (see below).
• Verstehbar makes no audio recording and does not archive captions on its servers. The event team can enable a local transcript.
• For pilot evaluation, we store aggregate counts. Stored context information and enabled technical diagnostics are described separately below.
• Reading along requires neither an account nor your name. Technically necessary connection data is still processed.
• For events created in our administration, a short link (vstb.s.gy/…) via the service Short.io may lead to the caption view. Opening it causes technical access data to be processed there (see below).

Who is responsible?

Responsible for this website and your contact for the Verstehbar service:

Popov Consulting, owner: Ilian Popov, Schiersteiner Straße 35a, 65187 Wiesbaden, Germany. Email: verstehbar@popov-consulting.de

When live captions are used at an event, the organiser decides on their use. The organiser is then responsible for the processing, and we provide the technology on their behalf (as a processor). You can still send us any questions; we forward them if necessary.

When you visit this website

Delivery via Cloudflare

This website and the caption views are delivered via the platform of Cloudflare, Inc. (101 Townsend St, San Francisco, CA 94107, USA). When you visit, Cloudflare processes technically necessary data: your IP address, date and time, the requested address, the amount of data transferred and information about your browser and operating system. This is needed to deliver the pages and to protect the platform against attacks and abuse.

The legal basis is our legitimate interest in a secure and working website (Art. 6(1)(f) GDPR). According to Cloudflare, technical logs of our server functions are kept for only a few days (3 to 7 days depending on the plan).

Cloudflare processes the data as a processor under a data processing agreement. Processing in the USA is possible; it is covered by the EU-U.S. Data Privacy Framework or the European Commission’s standard contractual clauses.

No cookies, no tracking

We use no cookies, no audience measurement, no advertising or social media services and no external fonts. No usage profiles are created.

Contact by email

If you email us, we process your address and the information you provide to answer your enquiry (Art. 6(1)(b) GDPR for enquiries about an event, otherwise (f)). We use an email provider as a processor. We delete messages once the enquiry is settled and no retention obligations apply.

When live captions are used at an event

Which data is involved?

• Audio from the authorised microphone and, where enabled, computer or tab audio, such as a Zoom session, plus the text produced from it. This can include remotely connected speakers. Text may contain names, personal information and sensitive content.
• Technical connection data of the devices reading along (for example the IP address when connecting).
• Anonymous counts to evaluate the pilot phase and the optional feedback “Did the captions help you?” (yes or no).

How it works

• The operator’s browser captures the authorised audio sources and sends audio encrypted directly to Soniox (Soniox, Inc., USA). Where needed, sources are mixed locally; computer or tab sharing uses only audio and does not transmit or store screen images. We use Soniox’s EU endpoints. Soniox states that audio and text are processed transiently in the EU, without storage or model training. Regional restrictions do not apply to system data such as account, usage and billing data.
• Text returns to the operator’s browser and is distributed via Cloudflare to the screen and devices reading along. The event rooms used for this are restricted to execution and storage in EU data centres. This restriction does not cover all Cloudflare services and technical metadata; processing outside the EU is possible for those.
• Our event server keeps a limited caption history in memory so that devices joining late see the current state. This history is not stored permanently on the server. An optional local transcript on the operator’s device is separate from it.

Optional local transcript

The event team can deliberately enable the transcript before or during a broadcast and stop it again. It is off by default and captures final recognised text and translations from the time it is enabled, plus technical information such as language, time and status changes. It can also contain text from preparation that has not yet been shown to the audience. No audio is recorded.

The transcript is stored locally in the operator device’s browser (IndexedDB) and can be exported as a text or JSON file. The stored transcript is not additionally sent to our servers, Cloudflare or Soniox; normal live text processing is unaffected. Data remains after closing the page or ending the event. There is no automatic deletion period. It can be removed in the operator view or by clearing the relevant browser data; exported files must be deleted separately.

Anyone with access to the browser profile or exported files can read the transcript. Operator credentials protect the interface, not locally stored data against access on the device. The organiser determines the purpose, recipients and retention of a transcript and informs affected people before creating it.

Stored context and names

To improve speech recognition, the event team can enter context, names and terms and save them per event on our server at Cloudflare. This information may contain personal data. Authorised operators of that event have access; it is not distributed to devices reading along or included in diagnostic journals. It is sent to Soniox for speech recognition when starting or deliberately applying the context. Saving alone does not trigger that transfer.

Stored context is not deleted when the broadcast stops. There is no automatic deletion period. The event team can clear both fields and save to remove their content from the current record. It should enter only necessary information and delete it after use. The limits explained below apply to server-side backups.

Anonymous counts and feedback

To evaluate the pilot phase, our server counts per event, for example, the highest number of devices connected at the same time, broadcast duration, interruptions and the yes/no answers. Only totals are stored, no content, no IP addresses, no device IDs. The numbers are deleted after the evaluation.

Short links via Short.io

For events we create in our administration, we generate a short link such as vstb.s.gy/K7M4R that redirects to the public caption view. The QR code, display and printed notice then show this short link. It is provided by the service Short.io (Short.cm Inc., USA). We only send Short.io the short code and the public address of the caption view, no content, captions or names.

When you open the short link, Short.io processes technically necessary access data: your IP address, date and time, the link opened and details about your browser, operating system and device type. Short.io stores this data, including the IP address and, where applicable, the previously visited page (referrer), as click statistics. We do not analyse these statistics. According to Short.io, statistics are kept for as long as necessary to provide the service; we delete the short link no later than together with the event's data.

Short.io operates data centres in the EU (Frankfurt) and in the USA; processing in the USA is possible. According to Short.io, transfers from the EU are safeguarded by the European Commission's Standard Contractual Clauses. We have requested a data processing agreement from Short.io. The legal basis is our legitimate interest in a short address that is easy to type (Art. 6(1)(f) GDPR). Without Short.io you can open the caption view via its direct link, which the event team will provide on request. Short codes are short and therefore guessable; they only lead to public caption views.

On your device

The caption view stores your language, font size, light or dark mode and whether you have already given feedback in your browser’s storage (localStorage). Stored preferences are not sent to us as a device profile; your selected language is taken into account for language-specific pilot evaluation. Local storage is required for the function you requested (Section 25(2) no. 2 TDDDG). You can delete it at any time in your browser settings.

Legal basis and recipients

The organiser determines the legal basis for using the service at an event, for example their legitimate interest in an accessible event (Art. 6(1)(f) GDPR). We only process the data on their instructions. Recipients are Soniox (speech recognition and translation, including supplied context information) and Cloudflare (text distribution and the configuration and operational data described here) as sub-processors. For short links, Short.io is an additional recipient (see above). A transcript is an additional processing activity; the organiser must also determine an appropriate legal basis for it and inform affected people beforehand.

Test mode: captions from online meetings (Recall.ai)

In a non-public test environment, we are trialling captions for an online meeting without a laptop at the venue. This only concerns our own Zoom test meetings with participants informed in advance, not regular events.

How it works

• A participant bot from the service Recall.ai (Recall.ai, Inc., USA) visibly joins the meeting when we start it. It only takes part with the meeting host's consent via Zoom's recording function.
• The bot sends the meeting's mixed audio encrypted to our server at Cloudflare. The server forwards it unchanged to Soniox (EU endpoints) and distributes the resulting text to the screen and reading devices, as at events.
• We use Recall.ai's EU region. Recall.ai does not store a recording: audio is only processed in memory (“Zero Data Retention”). No video, participant events or meeting metadata are recorded, and Recall.ai does not create a transcript.
• Our server stores neither audio nor captions nor the meeting link, only technical control data such as the bot ID, status, times and counters.

What data does Recall.ai receive?

Besides the transiently processed audio, Recall.ai receives the meeting link and the bot's display name. According to Recall.ai, it keeps the meeting link for up to 14 days after the bot finishes. Technical logs with request data including the meeting link are held by Recall.ai's subprocessors Svix (EU, 90 days) and Datadog (USA and EU, 7 days). Processing of this data outside the EU is therefore possible. According to Recall.ai, these logs contain no audio, text or participant details.

Legal basis and recipients

Taking part in the test meetings is voluntary; participants are informed in advance about the bot and this processing. The legal basis is your consent (Art. 6(1)(a) GDPR); you can withdraw it at any time with effect for the future. Recipients are Recall.ai (joining the meeting and transmitting audio), Soniox (speech recognition and translation) and Cloudflare (forwarding and text distribution). A data processing agreement with Recall.ai is in place; transfers to the USA are safeguarded by the EU Standard Contractual Clauses incorporated into it.

Technical operations and operators

Diagnostics and remote support

Where diagnostics are enabled, operator devices send technical operational states, fixed error codes, timestamps, counters and random session identifiers to Cloudflare. These support fault detection and remote assistance. Connection quality measurements, such as recognition delay and reconnections, may also be transmitted. This data is separate from aggregate pilot counts; we treat session identifiers as pseudonymous. The diagnostic journal contains no audio, captions, transcripts, context text or device names.

Diagnostics are active for enabled events in the pilot environment. Only the provider has access to the central history, using separate administrator credentials. The history is held in an EU-restricted event diagnostics store at Cloudflare. Retention periods are 30 days after the last observation for diagnostic records and 24 hours after the last report for current connection reports. Expired data is no longer used for display; storage is cleaned daily and on access. When the end of the view is reported, measurements are discarded; a technical identifier and sequence marker remain until cleanup so delayed reports cannot restore them. Exported diagnostic files are separate local copies and must be deleted separately.

Operator access and backup devices

For managed operator access, we store an identifier, an administrator-assigned label, enabled status, an optional validity window and creation and update timestamps per event. The access token is stored only as a hash. These records support access control and remain until administratively deleted; expiry or suspension does not delete the record. Labels should describe roles rather than contain unnecessary personal information. We also store, per operator access, the chosen order of the panels in the operator view (panel identifiers only, no content) so that it appears the same on every device using that access. It is removed when the order is reset or the access is deleted.

Where backup device coordination is enabled, random tab identifiers, chosen device labels, operational roles and audio input labels are processed. They are accessible only to authorised operators. Assignment ends on sign-out; coordination state is deleted when no participating connection or open incident remains. Orphaned state is discarded after twelve hours when next loaded. Tab identifiers and labels are also kept in the browser’s session storage. Active context is held transiently for takeover; deliberately saved context is separate from it.

Event management

When the provider creates events in the administration view, we store each event's public configuration in an EU-restricted database at Cloudflare (D1): event identifier, title, languages, display languages, presets for speech recognition (spoken languages, context, terms), status, planned end and creation, update and archiving timestamps. This information is publicly retrievable via the event link and may contain names if organisers provide them. We also log administrative actions with timestamp, action and event identifier only. For the short link we store its code, short address, the provider's identifier and processing status there, but no visitor data. Audio, captions, access tokens, IP addresses and device details are not stored there. An event's data is deleted 90 days after its end or, without an end date, 90 days after it is archived; this also removes the operational data stored for the event (operator access, settings, context, figures). Log entries are deleted after 90 days. For technical reasons the database keeps restore points for up to 30 days.

Backups and information for operators

Cloudflare may retain older versions of server-side storage in recovery backups for up to 30 days. Removal from the current record therefore does not immediately erase all backups. Operators must be informed about the operational functions used for their event. These functions do not change local transcript storage or the absence of server-side caption archiving.

Your rights

You have the right of access (Art. 15 GDPR), rectification (Art. 16), erasure (Art. 17), restriction of processing (Art. 18), data portability (Art. 20) and to object to processing based on legitimate interest (Art. 21). Just write to us at the address above.

You can also lodge a complaint with a data protection supervisory authority, in particular in the member state where you live or work or where the alleged infringement took place (Art. 77 GDPR).

There is no automated decision-making or profiling. You are not obliged to provide data; without the technically necessary connection data, however, we cannot deliver the pages.

Changes

Verstehbar is in its pilot phase and keeps evolving. If the way we process data changes, we update this policy. The version published here applies.